Exploring Decision Models in Neural Networks: Population Dynamics, Perceptual Decision Making, and Theory

Slide Note
Embed
Share

Dive into the world of decision models in neural networks with a focus on population dynamics and competition, perceptual decision making with V5/MT involvement, and the theory of decision dynamics including shared inhibition and effective 2-dim models.

bell_48
bell_48 Follow

Uploaded on Sep 22, 2024 | 0 Views

Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

E N D

Presentation Transcript

  1. Week 12 Decision models 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  2. Week 12-part 1: How do YOU decide?

  3. Week 12-part 1: Decision making turn Left? Right?

  4. Week 12-part 1: Review: High-noise activity equation noise model A (escape noise/fast noise) Population activity = ( ) ( ( )) F h t A t Membrane potential caused by input high noise = + d ( ) ( ) ( ) h t h t R I t dt = + + ext ( ) ( ) ( ) t ( ( )) d dth t h t R I w F h t ee w ee I(t) h(t) Attention: - valid for high noise only, else transients might be wrong - valid for high noise only, else spontaneous oscillations may arise slow transient ( ) A t = ( ( )) F h t

  5. Week 12-part 1: Review: microscopic vs. macroscopic (t ) An I(t)

  6. Week 12-part 1: Competition between two populations w w ee ee Input indicating right Input indicating left ( ) Ae 1 , t ( ) Ae t w , 2 ei w w ie ie (t ) Ainh

  7. Week 12-part 1: How do YOU decide?

  8. Week 12 Decision models 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  9. Week 12-part 2: Perceptual decision making? Is the middle bar shifted to the left or to the right?

  10. Week 12-part 2: Detour: receptive fields in V5/MT visual cortex IMAGE 1) Cells in visual cortex MT/V5 respond to motion stimuli 2) Neighboring cells in visual cortex MT/V5 respond to motion in similar direction cortical columns Albright, Desimone,Gross, J. Neurophysiol, 1985

  11. Week 12-part 2: Detour: receptive fields in V5/MT Recordings from a single neuron in V5/MT Receptive Fields depend on direction of motion Random moving dot stimuli: e.g.Salzman, Britten, Newsome, 1990 Roitman and Shadlen, 2002 Gold and Shadlen 2007

  12. Week 12-part 2: Detour: receptive fields in V5/MT Receptive Fields depend on direction of motion: = preferred direction = P

  13. Week 12-part 2: Experiment of Salzmanet al. 1990 coherence 0.8=80% coherence 0.5 = 50% coherence 0.0 Eye movement coherence -1.0 opposite direction Image: Salzman, Britten, Newsome, 1990

  14. Week 12-part 2: Experiment of Salzmanet al. 1990 Monkey behavior w. or w/o Stimulation of neurons in V5/MT Monkey chooses right N P fixation Visual stim. LED X = coherent motion to bottom right 1.0 -1.0 0.5 0.5 No bias, each point moves in random direction Blackboard: Motion detection/stimulation Salzman, Britten, Newsome, 1990

  15. Week 12-part 2: Experiment of Salzmanet al. 1990 coherence 0.8=80% coherence 0.5 = 50% coherence 0.0 excites this group of neurons coherence -1.0

  16. Week 12-part 2: Experiment of Salzmanet al. 1990 Behavior: psychophysics With stimulation

  17. Week 12 Decision models 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  18. Week 12-part 2: Experiment of Roitmanand Shadlenin LIP (2002) coherence 85% coherence 50% RF of Neuron in LIP: -selective to target of saccade -increases faster if signal is stronger - activity is noisy coherence 0% LIP is somewhere between MT (movement detection) and Frontal Eye Field (saccade control) Roitman and Shadlen 2002

  19. Week 12-part 2: Experiment of Roitmanand Shadlenin LIP (2002) Neurons in LIP: -selective to target of saccade -increases faster if signal is stronger - activity is noisy LIP is somewhere between MT (movement detection) and Frontal Eye Field (saccade control)

  20. Quiz 1, now Receptive field in LIP [ ] related to the target of a saccade [ ] depends on movement of random dots

  21. Week 12 Decision models, part 3 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  22. Week 12-part 3: Theory of decision dynamics = ( ) ( ( )) A t F h t activity equations n n Membrane potential caused by input ( ) h t = + + + ext ( ) ( ) t ( ( )) ( ( )) t d dth t RI w F h t w F h 1 1 1 1 ee ei inh = + + + ext ( ) ( ) ( ) t ( ( )) ( ( )) t d dth t h t RI w w F h t w F h 2 2 2 2 ee ei inh w ee ee Input indicating left movement Input indicating right movement ( ) Ae 1 , t ( ) Ae t w Blackboard: 2 , w ei ei population activity reduction from 3 to 2 equations w ie (t ) Ainh

  23. Population activity activity equations = ( ) ( ( )) A t F h t n n = ( ) (0) (1) 0.2 0.8 F h F F h for h = = 0.1 0.9 Inhibitory Population = = = + ( ) t ( ( )) t ( ) t ( ( ) t ( )) t A F h h w A A ,1 ,2 inh inh inh ie e e Blackboard: Linearized inhibition

  24. Week 12-part 3: Effective 2-dim. model = ( ) ( ( )) A t F h t activity equations n n Membrane potential caused by input ( ) ( ) h t = + + ) ( ( )) F h t ( ( )) F h t ext ( ) ( t d dth t h w 1 1 1 1 2 ee = + + ) ( ( )) F h t ( ( )) F h t ext ( ) ( ) ( ) ( t w d dth t h t h w 2 2 2 2 1 ee w ee ee Input indicating left movement Input indicating right movement ( ) Ae 1 , t ( ) Ae t 2 , w w ei ei population activity w ie (t ) Ainh

  25. Exercise 1 now: draw nullclines and flow arrows ext = + + d ( ) ( ) ( ) ( ) ( ( )) ( ( )) h t h t h t w g h t g h t 1 1 1 1 2 ee dt = ( (0) (0.9) (1) ) 0.2 0.8 g h g g g h for h (h ) g = 0.1 = 0.85 0.9 h = 1.0 = ext ext = = ; 5 . 1 = ; 8 . 0 1 h h w 1 2 ee ( ) ( ) h g h h h g h h 1= h 2= h d d 0 0 1 2 2 2 1 1 dt dt 1.0 0.8 0.2 0.0 1.0 0.8 0.2 0.0 Next Lecture at 10:36

  26. Week 12-part 3: Theory of decision dynamics Phase plane, strong external input 1= h d 0 dt = = ext ext 0.8 h h 1 2 = ( ) = 2 . 0 8 . 0 g h h for h ( 0 ) 1 . 0 g = ) 1 ( g 9 . 0 2= h d 0 dt

  27. Week 12-part 3: Theory of decision dynamics: biased input Phase plane biased input: Population activity ext ext h h 2 1 1= h 0 d ext 0 1= h d 1= 1= d d0 = 0 dt h dt h 2 . 0 h 2 dt dt ext = 2 . 0 h 1 2= h d 0 dt ext = 2 . 0 h 2

  28. Week 12-part 3: Theory of decision dynamics: unbiased weak Phase plane symmetric but small input 1= h d 0 ext 1 ext 2 = = 2 . 0 h h dt Weak external input: Stable fixed point 2= h d 0 dt

  29. Week 12-part 3: decision dynamics: unbiased strong to biased Symmetric, but strong input 1= h d 0 dt Phase plane 2= h 2= h 2= h d d d 0 0 0 dt dt dt

  30. Week 12-part 3: Theory of decision dynamics: biased strong Phase plane Population activity 1= h d 0 ext = ; 8 . 0 h dt 1 ext = 2 . 0 h 2 Biased input = stable fixed point decision reflects bias 2= h d 0 dt

  31. Week 12-part 3: Theory of decision dynamics: unbiased strong Phase plane 1= h d 0 dt ext 1 ext 2 = = 8 . 0 h h Homogeneous solution = saddle point decision must be taken 2= h d 0 dt

  32. Week 12 Decision models, part 3 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  33. Week 12-part 4: Review: unbiased strong Phase plane 1= h d 0 dt ext 1 ext 2 = = 8 . 0 h h Homogeneous solution = saddle point decision must be taken 2= h d 0 dt

  34. Week 12-4: Review: unbiased weak Phase plane symmetric but small input 1= h d 0 ext 1 ext 2 = = 2 . 0 h h dt Weak external input: Stable fixed point no decision 2= h d 0 dt

  35. Simulation of 3 populations of spiking neurons, unbiased strong input X.J. Wang, 2002 NEURON Popul. 2 Popul.1 stimulus w ei w w ie ie

  36. saccade onset Stimulus onset Roitman and Shadlen 2002

  37. Week 12 Decision models, part 3 12.1 Review: Population dynamics - competition 12.2 Perceptual decision making - V5/MT - Decision dynamics: Area LIP 12.3 Theory of decision dynamics - shared inhibition - effective 2-dim model 12.4. Decisions in connected pops. - unbiased case - biased input 12.5. Decisions, actions, volition - the problem of free will Biological Modeling of Neural Networks: Week 12 Decision models: Competitive dynamics Wulfram Gerstner EPFL, Lausanne, Switzerland

  38. Week 12-5: Decision: risky vs. safe How would you decide? goal

  39. How would you decide? goal Start

  40. fMRI variant of Libet experiment Decision and Movement Preparation -Subject decides spontaneously to move left or right hand - report when they made their decision Libet, Behav. Brain Sci., 1985 Soon et al., Nat. Neurosci., 2008

  41. What decides? Who decides? Your brain decides what you want or what you prefer but your brain this is you!!! -Your experiences are memorized in your brain -Your values are memorized in your brain -Your decisions are reflected in brain activities We don t do what we want, but we want what we do (W. Prinz) The problem of Free Will (see e.g. Wikipedia article) goal goal Start

  42. Decision, Perception and Competition in Connected Populations Wulfram Gerstner EPFL Suggested Reading: - Salzman et al. Nature 1990 - Roitman and Shadlen, J. Neurosci. 2002 - Abbott, Fusi, Miller: Theoretical Approaches to Neurosci. - X.-J. Wang, Neuron 2002 - Libet, Behav. Brain Sci., 1985 - Soon et al., Nat. Neurosci., 2008 - free will, Wikipedia Chapter 16, Neuronal Dynamics, Gerstner et al. Cambridge 2014

  43. Last Lecture in 2 Weeks: (holiday next week) - prepare questions for discussion section Exam: - written exam 17. 06. 2013 from 12:15-15:00 - miniprojects counts 1/3 towards final grade For written exam: -bring 1 page A5 of own notes/summary -HANDWRITTEN! The end

  44. Nearly the end: what can I improve for the students next year? Integrated exercises? Miniproject? Overall workload ?(4 credit course = 6hrs per week) Background/Prerequisites? -Physics students -SV students -Math students Comments: slides/notes

  45. now QUESTION SESSION! Questions to Assistants possible until June 4 The end and good luck for the exam!

  46. Exercise 2.1 now: stability of homogeneous solution )) ( ( ) ( t h g t A n n = Membrane potential caused by input ( ) ( ) ( ee h t b w = + + ) ( ( )) g h t g h t ( ( )) d dth t 1 1 1 2 = + + ) ( g h t ( ( )) g h t ( ) ( ) ( ( )) d dth t h t b w 2 2 2 1 ee ext 1 ext 2 = = Assume: h h b = = * ( ) h h h b a) Calculate homogeneous fixed point b) Analyze stability of the fixed point h(b) as a function of b 1 2 Next Lecture at 11:15

Related


Introduction to Neural Networks in IBM SPSS Modeler 14.2

Introduction to Neural Networks in IBM SPSS Modeler 14.2

anouk
Understanding Artificial Neural Networks From Scratch

Understanding Artificial Neural Networks From Scratch

inara
Graph Neural Networks

Graph Neural Networks

kornelia
Understanding Advanced Classifiers and Neural Networks

Understanding Advanced Classifiers and Neural Networks

elwynto
Exploring Neural Quantum States and Symmetries in Quantum Mechanics

Exploring Neural Quantum States and Symmetries in Quantum Mechanics

adelinda
Machine Learning and Artificial Neural Networks for Face Verification: Overview and Applications

Machine Learning and Artificial Neural Networks for Face Verification: Overview and Applications

lipp195
A Deep Dive into Neural Network Units and Language Models

A Deep Dive into Neural Network Units and Language Models

rowyn
Understanding Spiking Neurons and Spiking Neural Networks

Understanding Spiking Neurons and Spiking Neural Networks

kooper
Assistive Speech System for Individuals with Speech Impediments Using Neural Networks

Assistive Speech System for Individuals with Speech Impediments Using Neural Networks

albi
Understanding Neural Network Watermarking Technologies

Understanding Neural Network Watermarking Technologies

alad
Understanding Back-Propagation Algorithm in Neural Networks

Understanding Back-Propagation Algorithm in Neural Networks

kamilah
Exploring Biological Neural Network Models

Exploring Biological Neural Network Models

arielle
Understanding Network Analysis: Whole Networks vs. Ego Networks

Understanding Network Analysis: Whole Networks vs. Ego Networks

kayliee
Overview of Speech Recognition, Neural Networks, and Acoustic Models

Overview of Speech Recognition, Neural Networks, and Acoustic Models

sene_50
Understanding Mechanistic Interpretability in Neural Networks

Understanding Mechanistic Interpretability in Neural Networks

maksymilian
New Approaches in Learning Complex-Valued Neural Networks

New Approaches in Learning Complex-Valued Neural Networks

eliyahu
Computational Physics (Lecture 18)

Computational Physics (Lecture 18)

lorelai
Learning a Joint Model of Images and Captions with Neural Networks

Learning a Joint Model of Images and Captions with Neural Networks

cosimo
Biological Modeling of Neuronal Networks: Insights from Neural Dynamics

Biological Modeling of Neuronal Networks: Insights from Neural Dynamics

jellison_w
Understanding Keras Functional API for Neural Networks

Understanding Keras Functional API for Neural Networks

dale
Introduction to Deep Learning: Neural Networks and Multilayer Perceptrons

Introduction to Deep Learning: Neural Networks and Multilayer Perceptrons

ellie
Exploring Google's Tensor Processing Unit (TPU) and Deep Neural Networks in Parallel Computing

Exploring Google's Tensor Processing Unit (TPU) and Deep Neural Networks in Parallel Computing

enay_5
Exploring Noisy Output in Neural Networks: From Escape Rate to Soft Threshold

Exploring Noisy Output in Neural Networks: From Escape Rate to Soft Threshold

lharg
Exploring Membrane Potential Variations in Neural Networks

Exploring Membrane Potential Variations in Neural Networks

mair_s

More Related Content